Announcement

Collapse
No announcement yet.

DragonFlyBSD 6.0 Is Performing Very Well Against Ubuntu Linux, FreeBSD 13.0

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #21
    Vulnerabilities mitigations are in.

    Edit: checked. Mitigations are in DBSD since early versions of DBSD 5.x (~3 years)

    Conclusion: sour apples for some, whos claiming opposite.
    Last edited by aht0; 26 May 2021, 11:07 AM.

    Comment


    • #22
      Originally posted by aht0 View Post
      Vulnerabilities mitigations were AFAIK added long time a go. Including for newer ones, as new ones have been popping up now and then.

      Edit: checked. Mitigations are in DBSD since early versions of DBSD 5.x (~3 years)
      Yes, in FreeBSD since Feb 2018. They are largely identical to those in Linux, so there must be a better explanation for deterioration of Linux process scheduler (assuming Ubuntu is not misconfigured badly).
      Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

      Comment


      • #23
        Originally posted by Volta View Post
        ayumuNice fairy tales. Change Ubuntu governor to performance and say goodbye to bad results.
        That's amusing. You know, Dragonfly can also be tweaked for performance.

        What's interesting is that, with defaults, Dragonfly wins over what's probably the most common Linux distro.

        Linux is far from hitting a wall
        Call me back when Linux gets the sort of performance improvement Dragonfly did in the same timeframe it did.

        Linux has hit a wall, because getting any faster, even if just a little, takes a lot of work. Realize how well funded Linux is, and how small the Dragonfly team is.

        How is this possible? The difference is system architecture. No amount of money thrown at Linux will yield that sort of performance improvement.
        Last edited by ayumu; 26 May 2021, 11:34 AM.

        Comment


        • #24
          In this test of context switching It is also releveant how the test program is written.
          It is a user space program that is being tested/evaluated, then the linux kernel below it. Is the program written in C, perhaps with pthreads, or
          C++, perhaps with C++ stdlib ? Each provides its own peculiarities with regard to process/thread creation, handling, synchronization, etc, and how they translate into kernel constructs. It is more to consider here.

          Comment


          • #25
            Originally posted by jb.1234abcd View Post
            In this test of context switching It is also releveant how the test program is written.
            It is a user space program that is being tested/evaluated, then the linux kernel below it. Is the program written in C, perhaps with pthreads, or
            C++, perhaps with C++ stdlib ? Each provides its own peculiarities with regard to process/thread creation, handling, synchronization, etc, and how they translate into kernel constructs. It is more to consider here.
            Linux's IPC overhead is abysmal. This is a well-known fact. There's no point on trying to discredit the test, now.

            Comment


            • #26
              Originally posted by jb.1234abcd View Post

              Yes, in FreeBSD since Feb 2018. They are largely identical to those in Linux, so there must be a better explanation for deterioration of Linux process scheduler (assuming Ubuntu is not misconfigured badly).
              Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite
              If there was misconfiguration, it had to be caused by distributor. Michael is running benchmarks on default settings of any given OS.

              Comment


              • #27
                Originally posted by ayumu View Post

                Because FreeBSD chose to copy Linux in their approach to SMP (fine-grained locks).

                Matt used to be FreeBSD's technical lead, when FreeBSD was actually outperforming Linux, around 20 years ago. There was quite the argument as he'd reverted/rejected patches that simply went ahead with Linux's approach before a technical discussion on the matter could be held.

                It ended with Matt being ejected from FreeBSD. Some devs who also disagreed with the approach followed him to Dragonfly, which forked from FreeBSD.

                Slowly, with a small team, Matt's alternative approach, with a focus on low IPC cost and implementing lockfree/lockless system servers instead of locks, got implemented, furthered and to the point it is today, where Linux (and FreeBSD) are actually behind.

                Thus Matt's point can now be considered proven.
                In my view, the fact that he actually went ahead and both literally and figuratively put his money where his mouth is and invested his time and skills into creating DragonFlyBSD per his idea of the better approach compared to FreeBSD is worthy of respect in and of itself, no matter the outcome.

                The fact that this particular benchmark appears to validate his stance is merely a lovely bonus (inasmuch as it can be considered to validate your statement related to superior architecture). Hats off to him and the DragonFlyBSD team if this is not a fluke but a general and reproducible trend.
                Last edited by ermo; 29 May 2021, 11:20 AM.

                Comment


                • #28
                  An interesting fact I found yesterday is that Linux loses a lot of ground in some OpenMP workloads. The JtR test in particular scales quite badly beyond 4 threads on Linux whereas it seems to scale okay on BSDs. We'd need single-thread results from BSDs to confirm that but it seems like an interesting thing to investigate.

                  Comment

                  Working...
                  X